SDK_manager can't install CUDA

Autonomous Machines Jetson & Embedded Systems Jetson TX2
1

I use SDK_manager-0.9.13 to install cuda to TX2 by ethernet. But still have errors:

19:31:39 INFO : CUDA Toolkit for L4T : [target] [sudo] password for nvidia:
19:31:39 INFO : CUDA Toolkit for L4T :
19:31:39 INFO : CUDA Toolkit for L4T : Reading package lists...
19:31:39 INFO : CUDA Toolkit for L4T : Building dependency tree...
19:31:39 INFO : CUDA Toolkit for L4T : Reading state information...
19:31:39 INFO : CUDA Toolkit for L4T : E: Unable to locate package libfreeimage-dev
19:31:39 INFO : CUDA Toolkit for L4T : E: Unable to locate package libopenmpi-dev
19:31:39 INFO : CUDA Toolkit for L4T : E: Unable to locate package openmpi-bin
19:31:39 INFO : CUDA Toolkit for L4T : Unhandled error when running sudo apt-get install -y gnupg libgomp1 libfreeimage-dev libopenmpi-dev openmpi-bin
19:31:39 INFO : CUDA Toolkit for L4T : exit status 1
19:31:39 INFO : CUDA Toolkit for L4T : [ Package PostInstall Finished with Error ]
19:31:39 ERROR : CUDA Toolkit for L4T : NV_L4T_CUDA_TARGET_POST_INSTALL_COMP command scp -q -o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no /tmp/tmp_NV_L4T_CUDA_TARGET_POST_INSTALL_COMP.sh nvidia@192.168.2.208:~; ssh -t -q -o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no nvidia@192.168.2.208 "~/tmp_NV_L4T_CUDA_TARGET_POST_INSTALL_COMP.sh && rm -f ~/tmp_NV_L4T_CUDA_TARGET_POST_INSTALL_COMP.sh" finished with error
19:31:39 INFO : CUDA Toolkit for L4T :
19:31:39 INFO : CUDA Toolkit for L4T : [ 32.00 KB used. Disk Avail: 57.40 GB ]
19:31:39 INFO : CUDA Toolkit for L4T : [ NV_L4T_CUDA_TARGET_POST_INSTALL_COMP Install took 2m8s ]
19:31:39 INFO : CUDA Toolkit for L4T : command scp -q -o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no /tmp/tmp_NV_L4T_CUDA_TARGET_POST_INSTALL_COMP.sh nvidia@192.168.2.208:~; ssh -t -q -o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no nvidia@192.168.2.208 "~/tmp_NV_L4T_CUDA_TARGET_POST_INSTALL_COMP.sh && rm -f ~/tmp_NV_L4T_CUDA_TARGET_POST_INSTALL_COMP.sh" finished with error
19:31:39 ERROR : CUDA Toolkit for L4T : command terminated with error

I have tried these commands, but it doesn’t work.

sudo apt update
sudo apt --fix-broken install
sudo apt-get install libgconf-2-4 libcanberra-gtk-module python

I manually install libfreeimage-dev by ssh.but:

Reading package lists... Done
Building dependency tree       
Reading state information... Done
E: Unable to locate package libfreeimage-dev
2

Hi bwp2012,

Which JetPack version are you using?
Did you flash TX2 board before installing CUDA on it?
Can you please zip and attach the ~/.nvsdkm directory for more investigation?

3

For some special driver reason in kernel, I must use kernel of R28.2 version. But I want to use newest cuda. So I use SDK_manager-0.9.13 to config cuda-10.0 to kernel-4.4 of R28.2. So I meet this problem. But this process is too hard, so now I just use Jetpack-3.2.1 to config cuda-9.0.

4

CUDA releases from different JetPack releases are incompatible with each other. CUDA and installed o/s must be from the same release.